[Dev] Getting ready for run_tests

[Phillip J. Eby]

At 05:14 PM 6/6/2005 -0700, Ted Leung wrote:
>For those of you that don't know, PJE has written 'run_tests', which
>is a flexible driver for running the unit tests.   If you are
>interested in using it, you can type
>'RunPython -m run_tests' in $CHANDLERHOME and it will display its
>manual.   We're not quite ready to switch over to run_tests as the
>official method of running the tests (there are one or two issues
>that need to be ironed out), but it's there if you are interested in
>trying it.
>
>I have a small patch to RepositoryTestCase that turns of the
>repository commit log messages.  This makes running the unit tests
>under 'run_tests' a much more pleasant experience.
>
>I've attached the patch below, if no one objects, I'll commit it late
>tomorrow (Tuesday)

May I suggest logger.WARNING instead of critical?  The repository commit 
messages are at level INFO, so that will suppress purely-informational log 
messages without suppressing warnings or errors that might be useful for 
debugging.  I did a quick search for uses of the warning and error level 
messages in the repository.* packages, and the only use that was even 
remotely questionable was a warning about notifications taking longer than 
a second, that I've never seen happen in practice.

I'm also not sure if there need to be three places where the log level gets 
set; I think it's reasonable for RepositoryTestCase to set its own 
repository's logger level to WARNING, but I'm not sure the module should be 
messing with the root logger level.  Global log configuration seems more 
like a job for the test runner than the test cases.

By the way, I don't believe you need to have the 'stderr=False' 
change.  All that the 'stderr' flag does in the repository is tell it 
whether to add a dummy handler to the root logger.  It has no actual effect 
on the level of log suppression.

So, here's an alternative patch proposal.  It changes less, but still 
succeeds in suppressing the commit messages, and also allows individual 
test case classes to set their own log levels, if they want to raise or 
lower it for some reason:


Index: repository/tests/RepositoryTestCase.py
===================================================================
--- repository/tests/RepositoryTestCase.py      (revision 5605)
+++ repository/tests/RepositoryTestCase.py      (working copy)
@@ -20,6 +20,8 @@

  class RepositoryTestCase(TestCase):

+    logLevel = logging.WARNING      # a nice quiet default
+
      def _setup(self, ramdb=True):
          schema.reset()
          self.rootdir = os.environ['CHANDLERHOME']
@@ -46,14 +48,13 @@
              self.ramdb = False
              self.rep.open(ramdb=False,
                            fromPath=preloadedRepositoryPath,
-                          stderr=True,
                            refcounted=True)
+            self.rep.logger.setLevel(self.logLevel)
              self.rep.logger.info('Using preloaded repository')
          else:
              self.rep.create(ramdb=self.ramdb,
-                            stderr=True,
                              refcounted=True)
-
+            self.rep.logger.setLevel(self.logLevel)
              self.rep.loadPack(self.schemaPack)
              self.rep.loadPack(self.chandlerPack)
              self.rep.commit()


Of course, this doesn't do anything about the 186 print statements 
currently in Chandler and its tests (not counting commented-out print 
statements), so they can still produce quite a bit of spew.